Enabling LTE RACH Collision Multiplicity Detection via Machine Learning

TL;DR

This paper introduces a machine learning-based system for LTE RACH collision detection that not only identifies collisions but also estimates how many devices are involved, improving collision resolution and system efficiency.

Contribution

The paper presents a novel machine learning approach that estimates collision multiplicity in LTE RACH, enhancing collision resolution beyond existing methods.

Findings

Outperforms state-of-the-art preamble detection schemes

Accurately estimates the number of devices involved in collisions

Reduces RACH transmission latency and improves system load handling

Abstract

The collision resolution mechanism in the Random Access Channel (RACH) procedure of the Long-Term Evolution (LTE) standard is known to represent a serious bottleneck in case of machine-type traffic. Its main drawbacks are seen in the facts that Base Stations (eNBs) typically cannot infer the number of collided User Equipments (UEs) and that collided UEs learn about the collision only implicitly, through the lack of the feedback in the later stage of the RACH procedure. The collided UEs then restart the procedure, thereby increasing the RACH load and making the system more prone to collisions. In this paper, we leverage machine learning techniques to design a system that outperforms the state-of-the-art schemes in preamble detection for the LTE RACH procedure. Most importantly, our scheme can also estimate the collision multiplicity, and thus gather information about how many devices chose the same preamble. This data can be used by the eNB to resolve collisions, increase the supported system load and reduce transmission latency. The presented approach is applicable to novel 3GPP standards that target massive IoT, e.g., LTE-M and NB-IoT.